Spatial pooler and 2D input

[momiji]

Hi,

I’ve been trying to replicate / mess about with the MNIST work form the spatial pooler paper, and one thing that caught me off guard, the spatial pooler from as far as I can tell can expect a 2D input, i.e 28*28 bits in this case.

However when setup this way it seems to fail horribly, eventually leading to a loss in active columns and odd output SDR as seen in the figure.

Whilst if I setup the SP to have a 1D or flat input i.e 784 bits then it is capable of performing well and mapping to the input as shown in the other figure, and I’m curious as to why this would be?

[rhyolight]

This may or may not answer your questions, but it is a good introduction to the complexity topology introduces.

Thoughts about topology

I wanted to talk a bit about using topology with HTM. Most of this is just me working these thoughts out in my head, so some of it might be confusing or inaccurate. I welcome feedback and discussion. With no topology (the default today in NuPIC), the input space for the Spatial Pooler contains cells with no relationships to each other. The state of any bit in the input will never be correlated to the state of any other bit in the input. With topology enabled, the Spatial Pooler will calculate “…